Image Sampling Study - Chrome Material

[fade2blackened]

Hello all,

 

After reading both great 3DAS articles about vray (vray critical and image sampling in-depth), I decided to do a study myself.

 

This is actually because of an issue I´m having for quite a time.

When I have a thin chrome material, it gets very aliased. At the spot where the contrast is higher, it´s even worse.

 

I´ve made several tests and now I can´t really tell if it´s a GI or AA Sampling problem.

 

Well, let´s head to the images, they tell pretty much all I have tried:

 

1)Pretty basic settings, using a sharp filter (catmull). Some noise on the wall.

 

 

 

2)Decreasing the qmc min/max subdv. really messes up. Just to compare.

 

 

 

3)min/max still 1,1, but now with clr thresh 0,0

 

 

 

4)min/max now at 3/6. Both left and right "sticks" still very aliased.

to be continued...

[fade2blackened]

5)min/max = 1/4. A lot of alised edges and color issues there. Contrast is killing me!

6)Area filter now. Changed to adaptive subdvs -1/2. Get rid of the wall noise, but both the alised sticks are there.

7)Changed some GI settings (givind up on sampling!?). Not much difference, but the time.

8)A little tweak at the sampling with no lucky at all. Only time again.

to be continued...

[fade2blackened]

Is it a GI problem?

Is it a AA sampling problem?

Is it a AA filter problem?

Is it a resolution problem? (all my tests in a 640x480 res)

Is it a brain problem?

[fade2blackened]

Well...

Just a last try because I really need to go to bed:

 

Really high settings to a cropped render and a high render time for only a tiny part of the scene.

 

These are the settings I used:

-irrmap min/max=-3/1, hsph=50, interp=20, clr=0,3, nrm and dist=0,1

-lc 800 subdv

-800x600 res (before crop)

-adpt qmc min/max=1/100, clr=0,003, area filter (sharp filters really sucked)

-rqmc, adaptive amout=-0,5, noise=0,005, min samples=12, global mult=2,0

 

 

I KNOW most of the things I changed has nothing to do with what I wanted in the first place BUT I just wanted to make tests and I was freaking out already... lol

 

Here´s the result:

[ATTACH]19590[/ATTACH]

 

I think it´s better, but not CORRECT yet.

 

So now I ask...

 

Any help would be really nice!

[Crazy Homeless Guy]

if i understand the problem from what i have read in the past, you have to use sub-pixel mapping, and clamp the output in order to smooth out your highlight. ..basically, the highlight is several times brighter than the pixels next to it, and therefore the AA filters can not blend them properly to get a smooth result. if black = 0, and white = 1, then your highlight might register a 10. ...basically, the super bright area would probably be a glare if this were real life.

 

by clamping the output, it takes you out of floating point space, so it has drawbacks.

 

i read somewhere that there might be an insane solution with AA that might get them to render correctly, but there is not guarantee.

[Ricardo Eloy]

Well, Fábio, as Travis said, this is the famous floating point issue. The only way I know is to use a blur AA filter (you already know that) and clamping the output. Of course, one could argue that the size of your geometry might be of influence (a cylinder that thin), but this is something I have found myself struggling with for sometime. Then again, as Travis brought up, these really bright areas would come out as glares in the real life, so I normally add those in post. Not really noble, but...

[fade2blackened]

Hey Travis and Rick,

 

You´re both right about the sub-pixel mapping and clamp output settings.

After reading your answers I "googled" again (for the right thing now) and found some Vlado´s replies about this issue:

 

The correct way to deal with this issue is to turn on the "clamp output" and "sub-pixel mapping" options in the Color mapping section - you will get better AA, but a somewhat wrong rendering.

 

The problem is intrinsic to rendering HDR images. There are only two things that you can do about it: either force the renderer to clamp the colors (or map them in another clamping way) and lose all HDR information, or blur overbright objects to hide the jagged edges - which is what happens with real-world cameras.

 

Note that unless the renderer clamps the colors, you will not be able to remove the jaginness no matter how many AA samples you put into it. On the other hand, once you clamp the colors, you lose all the HDR information - but you get nicer AA in these situations.

 

 

So...

I guess this leads to a post production "re-work" rather than trying to fix this at render time.

 

Is it done in most cases?

 

 

I´ve made new renders so I can compare again. (attached)

When I turned both sub-pixel and clamp on, the image got really messed up, so I don´t think that´s the best option again.

[ATTACH]19601[/ATTACH]

 

 

 

Thanks for the replies.

[]